A technique is described for automatically determining the loss of material in an engineering component due for example to wear or corrosion. The system is based on the subtraction of two digital images of two contour maps of the surface of the object in its worn and unworn state. Calibration tests give an accuracy within 5%.

We present a scanning triangulation system for large objects with high lateral and longitudinal resolution. The system is based on axicons in the illumination device and on the use of the Scheimpflug-condition in the observation path. We describe a scanning-by-deflection geometry that minimizes the number of necessary detector pixels.

The suitability of a fast, short range time-of-flight distance sensor for industrial applications is considered. Critical characteristics of the sensor are evaluated and compared to those of other type of sensors. Results obtained with a few experimental devices are given. At a range of a few meters the speed of measurement can be up to 2.10⁶ measurements/s with a single pulse resolution of the order of 10 mm. Using averaging the resolution can be improved down to the level of about 100 pm. An example of the design of a fast, short range time-of-flight distance sensor is given. The sensor described is designed to be used at short ranges (<5 m) for passive targets mainly in 3D image acquisition applications. The construction described gives a good concept of the technique needed and its suitability for industrial use.

A new method for measuring the flight time in high speed, high resolution ranging applications is presented. The detailed theoretical consideration of the method shows that the method is well suited to these applications. Examples of practical constructions and their performance show that the method can be put into practice with a very small number of components with good results.

Real-time, non-contact, high accuracy measurements of small components are realised utilizing diffraction pattern theory. High speed line-scan CCD (charge coupled device) sensors, coupled with a microcomputer system with specalised hardware, allow the implementation of least squares and transform algorithms for the analysis of diffraction pattern data.

An optical triangulation system using television cameras measures lineside structures to ensure adequate clearances for railway rolling stock. The measuring system is mounted on a railway vehicle and uses white light illumination. Operation is possible at speeds of up to 100 km/h which allows complete routes to be measured in a single shift. Results Ire referenced with respect to the railway track and are recorded on digital magnetic tape.

Institut CERAC has been involved in optical metrology and 3-dimensional surface control for the last couple of years. Among the industrial applications considered is the on-line shape evaluation of machined parts within the manufacturing cell. The specific objective is to measure the machining errors and to compare them with the tolerances set by designers. An electro-optical sensing technique has been developed which relies on a projection Moire contouring optical method. A prototype inspection system has been designed, making use of video detection and computer image processing. Moire interferograms are interpreted, and the metrological information automatically retrieved. A structured database can be generated for subsequent data analysis and for real-time closed-loop corrective actions. A real-time kernel embedded into a synchronisation network (Petri-net) for the control of concurrent processes in the Electra-Optical Inspection (E0I) station was realised and implemented in a MODULA-2 program DIN01. The prototype system for on-line automatic tolerance control taking place within a flexible machining cell is described in this paper, together with the fast-prototype synchronisation program.

Equipment for automated inspection is becoming increasingly available in fully developed commercial form. Trends can be seen emerging which allow such equipment to he applied without significant application-specific development work by the end user or supplier. Areas of particular concern to non-expert end users are identified, based on the author's experience gained as an independent consultant in the field of machine vision, with no commercial ties to any manufacturer or supplier. Examples from the extreme ends of the cost spectrum are quoted, and the relative lack of commercial solutions to three dimensional information requirements is noted.

Many equipments for automatic inspection with vision have been incorporated in production lines of nearly all car manufacturers. RENAULT also has now a three years of experience with automated vision and some rules have been established. Our most important contributions have been : - Examples of applications, some now operating, some waiting for integration in complete systems. - How to establish a good "request to quote" ? - How to examine and compare suppliers'offers ? What selection criterias and important questions to ask for ? - What can be expected from the new vision equipment and what are the needs in hardware and software.

A brief review of the theories used to solve the problems of characterization of rough surfaces from optical measurements is made. A special attention is paid to the very important case of microrough surfaces, where the order of magnitude of the width of the irregularities is less than ten microns. We compare the limits of validity of the main theories : theory of speckle developed in Optics, approximate electromagnetic theories and rigorous electromagnetic theories. We show that, with the recent improvements of the rigorous theory, very interesting numerical results can be obtained, even though the numerical limitations cannot be neglected. For example, we use rigorous numerical results to show that two random microrough surfaces having the same correlation function, illuminated by the same incident beam, can generate diffraction patterns at infinity having very different statistical properties.

Scratch quality assessment is an emotive subject between users and suppliers with the topic being far more complex than one would imagine. The ultimate objective must be to be able to determine if and by how much scratches affect optical performance. However before this topic can be addressed with confidence objective methods of scratch measurement need to be established in order that organisations can agree on the relative severity of individual scratches. An important element in this process is the definition and production of standard scratch artefacts which will provide the vehicle for assessment either by visual or machine assisted means. This paper reviews a comparative experiment which set out to correlate existing British Standard scratches with those used in the United States in conjunction with MIL-0-13830A. The results of the experiment showed considerable inconsistency between operators regarding relative severity of the sample scratches and poor correlation between sets of supposedly standard scratches. The visual test was followed by a machine assisted evaluation which removed the subjective discrepancies between individual operators but confirmed the inconsistencies within National Standard scratch artefacts. Until such time that scratch quantification can be achieved by optical performance alone, machine assisted methods and good standard artefacts need to be developed which provide results which are consistent with manual use of artefacts.

The characteristics of surface required to ensure fitness-for-purpose depend greatly on the particular functions to be performed by that surface. Although still widely used, the electroniccontacting stylus leaves much to be desired in terms of ease of use, area coverage and cost. Alternative means for assessing surface texture including roughness, waviness and flaws have been widely studied in recent years in an attempt to produce a measuring system in keeping with the requirements of modern manufacturing. A review of the methods of operation of optical sensors, including an account of their principal advantages and disadvantages, will be presented together with an indication of how recently developed methods might overcome some of the application problems.

For objective quality control of paint coatings a videooptical method has been developed and realized in a prototype system, which consists of an optical sensor with a standard CCD camera and a digital image processing unit controlled by a microprocessor. The system allows a colour independent quantification of surface quality in less than one second.

Developments towards an automatic inspection system based on Electronic Speckle Pattern Interferometry (ESPI) are presented. Emphasis is placed on the techniques for the automatic analysis of the Speckle Pattern correlation fringes to produce three dimensional surface profiles from which various physical parameters can be derived. The range of application and general system design principles will be discussed.

Recent developments in digital processing have renewed interest in E.S.P.I., whereby direct contouring is possible with fringes of interferometric quality. It is now possible to locate points on the object accurately in association with E.S.P.I. analysis, although a trade-off between point resolution and fringe visibility seems likely. Nevertheless the time is close for the development of a complete optical inspection system.

A hologram of an intricate mechanical assembly can be used to project a three dimensional image which can be examined optically. Any part of the image can be observed provided that the corresponding part of the mechanical assembly was illuminated during the construction of the hologram and was visible from any part of the 'window' formed by the holographic plate. Accurate reconstruction from a carefully processed holographic plate yields an image which is within ± 1% of the size of the original object. When measured markers are included with the object, adjustments can be made to make the reconstruction more accurately 1:1 and image accuracies approaching 0.1 % have been achieved. A three axis computer controlled measurement system is employed to move a television camera throughout the image volume. The accuracy of the system is better than 5pm and the repeatability of the measurements is 50 μm within a volume measuring 600 mm x 300 mm x 300 mm. The television camera utilises an 18 mm image tube which gives a monitor image representing an area of 8.7 mm by 6.7 mm with each scan line representing a distance of 11.6 μm. The form of any part of the mechanical assembly is known and a processing system can produce a mask to compare with the acquired image. The difference between the two can be evaluated and the camera position can be changed until the difference is a minimum. The image is adjusted to be within 1 pixel of the mask and accurate measurements of the position, in three dimensions, of selected parts of the structure determined in this manner. The paper will include an analysis of the causes of loss in measurement precision and discuss modifications which could lead to improved performance.

Vision systems can be used for a wide range of tasks within research and industry. Historically such systems have been used for non critical inspection. Increasingly these systems are now used for measurement, often to engineering tolerances. The requirement for calibration with vision is discussed and a solution particularly relevant to automatic processes presented.

Humans are endowed with the facility to perceive colour. This not only provides an additional aesthetic dimension but also helps perform visual tasks efficiently. There are many occupations, including inspection, not open to those with defective colour vision. Todays machine vision systems are virtually all colour-blind. Yet there are applications where colour is intrinsic. Consider for example the inspection and grading of fruit, vegetables, biscuits and other food products. Consider also the widespread use of colour coding for wiring and components in the electrical and electronic industries. Automatic optical inspection of such things cannot be done without relating to colour. There are other applications where colour is not directly relevant but the additional information provided can help simplify and speed up the processing task. This paper reviews the nature of colour, relating the psychophysical aspects of colour perception and the physical properties of available sensors to the needs of an automatic inspection system. The theory of colour perception is based on the tri-stimulus theory which says that any colour may be matched using appropriate proportions of three primary colours. Although later experiments have suggested human colour perception is more complex, most electronic video sensors employ a three colour system. Usually the red, green and blue primary components are derived and may be used directly as sensory inputs to a vision system. However the primary representation of colour is not the most efficient means of encoding nor is it the most useful basis for interpretive processing. The R,G and B primary signals may be simply transformed into a new coordinate system where one of the axes represents true object colour or hue. Using this new colour space simplifies processing. These ideas are illustrated by an inspection example. The colour coded wires of a European power cable are identified to ensure that a power plug is safely wired. For this application a straightforward and reliable inspection system can only be produced using colour information.

Machine vision systems are rapidly becoming the standard way of automating routine visual tasks such as inspection and reading. The systems operate by capturing an image, converting the image into bits of data, and processing that data. In converting an image into data, early machine vision systems simplified the image into one with only two intensity levels--black and white. Such "binary" systems thus discarded much information. More advanced machine vision systems process all of the shades of grey present in an image and consequently have more information on which to base their decisions. This paper explains how grey-scale and binary processing differ. It shows that for accurate, consistent performance, true grey-scale systems are superior.

Smith Associates is automating the operation of a contact lens factory. The latest stage of this project involved the development of an automatic radiuscope to measure the properties of the back face of the lens. The equipment developed was based on an existing manual radiuscope. A solid state video camera was installed in place of the eyepiece, and stepper motors used to drive the manual controls. An image processing computer, derived from Smith Associates' PIX system, was used to carry out the image processing and control the stepper motors. The equipment has been in service under factory conditions since mid 1985. It has been found that the measurements made by the equipment are about an order of magnitude more accurate than those made by a person using the same optical system. Also, a number of measurements are made which could not be made manually. These include surface finish, toricity and eccentricity of the back face of the lens.

The introduction of flexible manufacturing systems and the increased use of CNC/DNC machines creates a high capacity in a small batch production environment. As inspection is always an after event, it is usual in this environment to accurately measure the first off machined sheet, prior to the manufacture of the remainder of the batch. Assuming that the machine parameters, instructions, tool selection and use do not change, then the remainder of the batch should be made consistent with the first off inspected "good" item. Unless the whole batch is to be manufactured at risk, the CNC or DNC machinery is prevented from completing the remainder of the batch until the first item is satisfactorily inspected. Therefore this inspection must take place quickly in order to minimise the elapsed time factor and the loss of production capacity. Additionally, it is extremely difficult for an inspector to measure, check tolerances and positional accuracy of a multiplicity of shapes cut in a large area sheet, even when assisted by such conventional aids as touch probe coordinate measurement machines. Such machines are also not very applicable to the inspection of small features cut in thin material. Figure 1 shows a typical work-piece pattern which can be stepped and repeated across the total area of the sheet material, prior to being cut into its constituent parts. This paper describes the development of a Vision System for Automatic Measurement purposes and the initial experiences of its use as an inspection tool.

In many industrial inspection applications, the high speed analysis of greyscale images is required. ICOS has developed vision systems, which can be, because of their programmability, used in a wide variety of vision applications. Four examples of applications in the electronics industry are presented and the main features of each application are described. The first application deals with the inspection of reed switches. The second and the third application are an inspection and alignment task in the assembly of PCBoards. The last application is a combined alignment and inspection task in the semiconductor industry. These present a sample of industrial applications solved with ICOS vision systems.

The fabrication of very large scale integrated circuits, usually done on silicon substrates known as wafers, is a process by which successive layers of films are deposited and patterned. Photolithographic transfer of patterned images is performed fEom a master "reticle" with a step and repeat camera, with imaging Fspabilities of -2 x 10⁸ discrete 1 elements in a single exposure of a field area of 2cm . Typical processes contain approximately 60 exposure fields per wafer level and have UP to 15 discrete levels. Each level requires its own reticle pattern. The number of devices printed on a single wafer varies but is usually between 50 and 1000 depending on the type of device and the wafer size. Even for successful processes, typically only 50-75% of available devices actually are functional at completion. Loss of these devices is primarily due to defects encountered in the processing. It is straightforward to create a first order estimate f the nominal defect density requi5ed to support such yields by using Poisson statistics. Assuming a device size of 0.5cm ' and 10 critical levels, the defect density required to give 60% die yield is given by 2: -nDoA Yield =e 1 where n = critical device levels (1) 1 + nAD A = area of device 0 Do = defect density per wafer Do = 0.13 defects/cm2 per critical level Since device killing defects can be smaller than 1.0um, the ability to discover such defects and remove the source causes of defects is critical.

Within the last decade, machine vision has emerged as a practical manufacturing tool in many industries. The electronics industry has found a particularly large number of applications for the technology, including automatic inspecting, reading, and guidance. This paper gives a brief overview of the varied applications of machine vision in electronics. It then proceeds to analyze one application in greater detail - automatic inspection of electronic marks.

A novel fibre optic illuminator with a spatially variable output light intensity distribution was built. By the aid of a semitransparent mirror and lens optics, this light distribution was projected onto the object to be imaged directly from the direction of observation. It was found that very homogeneous intensity distributions can be achieved in the image plane of the camera by compensating the cosine-fourth law of the camera optics. This allows for a constant, position independent threshold setting for digitizing the image. We will report on very favourable results obtained for the illumination of hybrid circuit boards and wafer surfaces. The influence of object size, surface quality (diffuse or highly reflective) and polarized light will be discussed.

The measurement of the size of moving holes is described by processing their image using spatial filtering and converting their size to electrical signal. In the present case, the shape of the hole was assumed to be circular and the band-pass filter giving high sensitiv-ity was designed. In deciding the aperture size for the photodetector, tolerance to position fluctuation of the moving holes was taken into consideration. Dealing with the holes of various size within 30011m ±10% diameter, the accuracy obtained was within l%,i.e. 3μm. Moving speed of the holes was 20cm/sec and their position fluctuation was allowable near to their size.

In the process of continuous spinning it is particulary important the measurement of the velocity of textile fibers: in fact the uniformity, smoothness and tenacity of the yarn depend on the velocity of the fibers at the moment they reach the twisting organ and on the average number of the collected fibers. It is important that the applied tecnique should not affect the flow. An electro-optical system is described for the measurement of the velocity of fibers ( 20 mm long, 25 μm diameter) suspended in an airflow. The samples are carried along by the surrounding medium in a duct that has two windows (one for lighting, the other for viewing). A cross correlation of the detectors output is carried on to obtain the velocity and density of the fibers.

A real-time Vander Lugt correlator for the automatic recognition, location and inspection of objects from direct imaging or TV camera input is described. Techniques to cope with changes due to scale and rotation are discussed. These include: (i) Image rotation with dove prisms. (ii) Multiple matched spatial filters by parallel addressing. (iii) Selective spatial filtering. Example results from an inspection task are presented.

A system is described which allows for detection, recognition and classification of various types of defects in textile slivers. Three detection units with different viewing conditions are used, comprising linear photodiode arrays.

The Ramer reasons for development of new in-line sensors for food quality, and includes examples of optical sensors for in-line measurement and automatic control. The development is described of a colour sensor, the Qual-Probe, in which illumination from modulated single wavelength sources (LEDs and IREDs) is projected towards a small region of food material while it is in motion during the manufacturing process. One or more detectors are used to collect the diffusely reflected light from the controlled sources, by methods which are self-compensating for changes in source power or ambient radiation.

The problem. often, arises in production processes of identifying work-pieces by means of an identification number or other alpha-numerical symbols, for the purpose, for instance - of control of a multiple-branched material flow, - of supervision of progress in manufacture, - of individual work-piece-specific storage of quality data in a quality monitoring system. Various systems can be used for this purpose. One can make use of traditional methods, using cards which indicate work orders and further informations. On the other hand, mechanical codes or microwave-based information-units can be affixed to the transportation system (e.g. pallet) for the purpose of automatic identification. Bar codes nowadays have a dominant position in the field of encoding of information in automated material-flow and manufacturing systems.

A description is given of an image processing system under development at UWIST. This is capable of a throughput of up to ten automated visual inspection tasks per second. When fully developed it will be a turnkey system for the factory floor. Running under control of an MC 68000 based host, it uses a linear array processor (LAP) designed at the UK's National Physical Laboratory (NPL). Software has been produced at the high and intermediate levels, and timing comparisons of image processing algorithms are made with two serial processors, the PDP 11/23 and the MC 68000.

This paper describes a completely automated chromosome analysis system. It comprises of an automated microscope, an image analysis system for evaluation of the microscopic fields, and peripheral equipment for user dialogue and control of the procedures. Photo-quality hard copies of the completed karyograms are output via a special printer. Metaphases are found fully automatically by the system. A research quality microscope is used for automatic screening of up to 16 slides in one metaphase finding run. The metaphase finding is based on a special cluster analysis. By resetting the metaphase models, the system can be adapted to different staining techniques. Metaphase coordinates are stored in the system enabling automatic repositioning under high magnification. The system can be used for counting the chromosomes of all metaphases as well as for karyotyping, based on banding analysis. The karyotyping algorithm is based on classification of the banding pattern of each chromosome. The knowledge base is derived from a banding pattern analysis of 4000 chromosomes.

The usefulness of a lens depends on its imaging performance over its entire field of view. An effective test must measure image quality at off-axis positions, as well as on axis. An automated system is described which quantifies image quality at several different field positions more quickly and precisely than is possible with a manual system. In addition to the Modulation Transfer Function (MTF), the Phase Transfer Function (PTF), field curvature and lens distortion can be measured. Algorithms used to find the off-axis image are discussed in detail, including trade-offs between speed and generality. It is shown that a well designed algorithm must take into account the expected characteristics of the off-axis image, with allowances for uncertainties. The operation of the test system is illustrated with examples.

A problem-oriented color measuring system has been designed for the design of optical sorting systems relying on color discrimination. High performance and efficiency can be obtained in industrial applications.

High-quality polymer production depends on thorough and uniform mixing of the raw materials. Effective research on the mixing process requires precise, repeatable, and objective measurements of "mixing quality". The accepted technique for measuring mixing quality in the early stages of the mixing process is to load the extruder with raw materials in two contrasting colors, and to extract samples of the "mix" at various points along the extruder. Photographs of these samples are then evaluated visually to determine the statistics of the length and width distributions of the striations produced by the two colors. Direct methods of performing this analysis are highly subjective and are extremely labour-intensive. An image analysis technique that mimics the direct measurement method and computes the length and width histograms is presented. The technique uses an industrial vision development system, which carries out a grey-level-encoded skeletonizing process. Examples of processed images and results are included.

A new computer-aided system for automatic fringe analysis will be presented. This commercially available system is based on the phase-shift method. It enables the complete analysis of fringe patterns for various optical measuring techniques: 2-beam interferometry, speckle-interferometry, holography, photoelasticity, fringe-projecting methods, and Moire techniques.

Rapid development of reliable high power lasers, of an instant processing of photothermoplastic film and of a computer based fringe interpretation system enables the application of holographic interferometry in industrial environment. Holographic interferometry offers some essential advantages - non contact, inertialess measurement - the object is not affected - 2-dimensional image like information - testing under practise related loading - relevance between holographic indication and actual flaws The key for industrial application is a computerbased fringe interpretation.

The role of Artificial Intelligence (AI) in industrial vision systems is examined. It is concluded that there are several, quite distinct areas where AI concepts and techniques are likely to be useful. The first is that of designing and planning industrial vision systems. The most obvious application is probably that of analysing complex scenes, about which little or nothing is known in advance. However, the author argues that in a well designed industrial plant there is very little disorder. For example, the orientation of piece parts would not be lost by tossing them into a bin. The cost of re-orientating them can be quite expensive. The well known "bin picking" problem is thus seen as one for which solutions should be sought but instead avoided. The third area where AI techniques might be used is in the task of inspecting either complex objects / assemblies, or those which are made in small quantities. The paper concludes with a discussion about the requirement for a convenient language for expressing both image processing and AI algorithms. The structure of such a language is described.

The aircraft design process is characterized by the application of a wide range of knowledge, based upon a certain degree of judgement and experience of the designer. In this paper we outline this process and the scope of a computer program which performs one aspect of the overall task related to wing design. Emphasis is given to the contrast between wing design and the more elaborate requirements needed to describe the aircraft configuration. It is shown how the solution of the latter problem is aided by identifying different types of knowledge.

A method of recognising a simple object in a cluttered scene is described, which uses geometrical models to express detailed knowledge about the object and the scene. Salient features of the object serve as cues which guide a search for an appropriate instance of the model; a pre-processor identifies a minimal set of tests for the cue. The method is demonstrated by a program which detects London buses in street scenes, but may be applied to a wide variety of other visual tasks.